t

app.py

@@ -194,19 +194,38 @@ def f(context, temperature, top_p, max_length, model_idx, SPACE_VERIFICATION_KEY
 
             # TODO use fallback gpt-2 inference api for this as well 
             # TODO or just make it an option in the menu "GPT-2 inference"
-        else: 
+        elif model_idx == 2: 
             DISTIL_GPT2_API_URL = "https://api-inference.huggingface.co/models/distilgpt2"
-            generated_text=""
-            while (max_length > 0):
-                # NOTE see original implementation above for gpt-J-6B
-                payload = {"inputs": context, "parameters": {"max_new_tokens": 250, "temperature": temperature, "top_p": top_p}}
+            # generated_text=""
+            # while (max_length > 0):
+            #     # NOTE see original implementation above for gpt-J-6B
+            #     payload = {"inputs": context, "parameters": {"max_new_tokens": 250, "temperature": temperature, "top_p": top_p}}
+            #     response = requests.request("POST", DISTIL_GPT2_API_URL, data=json.dumps(payload), headers=headers)
+            #     context = json.loads(response.content.decode("utf-8"))
+                    
+            #     context = get_generated_text(context)
+
+            #     generated_text += context
+            #     max_length -= 250
+            generated_text = ""#context #""
+            while len(generated_text) < max_length:#(max_length > 0): NOTE NOTE commented out this line and added new check
+                # context becomes the previous generated context
+                # NOTE I've set return_full_text to false, see how this plays out
+                # change max_length from max_length>250 and 250 or max_length to 250
+                payload = {"inputs": context, "parameters": {"return_full_text":False, "max_new_tokens": 250, "temperature": temperature, "top_p": top_p}}
                 response = requests.request("POST", DISTIL_GPT2_API_URL, data=json.dumps(payload), headers=headers)
-                context = json.loads(response.content.decode("utf-8"))
+                context = json.loads(response.content.decode("utf-8"))#[0]['generated_text']
+                # context = get_generated_text(generated_context)
+                
+                # handle inconsistent inference API
+                # if 'generated_text' in context[0]:
+                #     context = context[0]['generated_text']
+                # else:
+                #     context = context[0][0]['generated_text']
                     
-                context = get_generated_text(context)
+                context = get_generated_text(context).strip()
 
                 generated_text += context
-                max_length -= 250
 
                 # payload = {"inputs": context, "parameters":{
                 #     "max_new_tokens":max_length, "temperature":temperature, "top_p":top_p}}
@@ -214,6 +233,19 @@ def f(context, temperature, top_p, max_length, model_idx, SPACE_VERIFICATION_KEY
                 # response = requests.request("POST", API_URL, data=data, headers=headers)
                 # generated_text = json.loads(response.content.decode("utf-8"))[0]['generated_text']
             return generated_text#context #_context+generated_text
+        else:
+            url = "https://api-inference.huggingface.co/models/gpt2-large"
+
+            generated_text = ""#context #""
+            while len(generated_text) < max_length:
+                payload = {"inputs": context, "parameters": {"return_full_text":False, "max_new_tokens": 250, "temperature": temperature, "top_p": top_p}}
+                response = requests.request("POST", url, data=json.dumps(payload), headers=headers)
+                context = json.loads(response.content.decode("utf-8"))
+                context = get_generated_text(context).strip()
+
+                generated_text += context
+            return generated_text
+
 
     except Exception as e:
         return f"error with idx{model_idx}: "+str(e)